We now examine IAV in June 1-September 15 (JJAS15) daytime mean ozone deposition velocity (vd). We also examine diurnal cycles to probe the mechanisms that control vd. We find IAV in the shape of the JJAS15 mean diurnal cycle (Figure 3.2a): some years have
a broad daytime maximum (1992, 1996), while others have a morning or afternoon peak (1993, 1995 versus 1997, 2000).
JJAS15 daytime mean canopy resistance (Rc) exceeds both the aerodynamic and quasi-laminar resistances by roughly an order of magnitude (Figure B.2), confirming that Rc controls IAV in JJAS15 daytime vd. We investigate how stomatal and nonstomatal con-ductance (gs and gns, respectively) impact IAV in Rc using two process-level gs models driven by Harvard Forest observations and derived parameters (e.g., GPP) (Section 3.2.3).
During 1992-2000, JJAS15 daytime mean gs spans 1.1-1.6 cm s−1 for the Lin et al. (2015) model (L15) and 0.37-0.54 cm s−1for the Shuttleworth et al. (1984) inversion of the Penman-Monteith equation (P-M). Figure 3.2b shows JJAS15 daytime mean gsfor 1992-2000 (values for each year are normalized by the respective multiyear means, which are shown in cm s−1 in black asterisks). Despite the discrepancies in magnitude, both estimates consistently show little IAV: the relative interannual spread (i.e., relative standard deviation, or coef-ficient of variation, across yearly values) of daytime mean gs for 1992-2000 is ±10.9% for P-M and ±13.8% for L15. Neither approach to estimating gs yields a ranking of low to high years that matches the gs ranking (e.g., high gs years are not high vd years and low gs years are not low vd years) (Figure 3.2b,c). The similarities in IAV between P-M and L15 suggest that gs does not control IAV in JJAS15 daytime mean vdwith the caveat that these models may not be sufficiently sensitive to low-frequency environmental controls.
For each gs estimate, the shape of the mean diurnal cycle for most years deviates little from the shape of the multiyear mean diurnal cycle (Figures 3.2d, B.3b,d). Assuming that existing gs models can adequately represent IAV in the gs diurnal cycle, we conclude that deviations from the shape of the climatological diurnal cycle of vd(Figure 3.2a) are largely controlled by gns.
We place more confidence in P-M than L15 for the following reasons. First, a recent finding using isotopic methods suggests that standard partitioning between daytime ecosys-tem respiration and GPP overestimates GPP by 25% during June-July at Harvard Forest (Wehr et al., 2016). This finding may challenge the assumptions inherent in photosynthesis-based approaches to estimate gs (e.g., L15). Second, the magnitude of P-M gs is less than gc (Figure 3.2d) so P-M gs can be accommodated in our resistance-in-series framework for
L15 P-M
Figure 3.2: (a) June 1-September 15 (JJAS15) hourly mean ozone deposition velocity (vd) from observations at Harvard Forest. Black indicates the multiyear mean. Error bars indicate two standard errors. JJAS15 daytime (9am-3pm) mean (b) L15 and P-M stomatal conductance (gs) for each year, normalized to the respective multiyear mean (the multiyear means are shown in asterisks in cm s−1), and (c) canopy conductance (gc), and nonstomatal conductance (gns) (cm s−1) and the percentage stomatal contribution to gc(gsg−1c ; %); the latter two are inferred using P-M gs. (d) JJAS15 multiyear hourly mean gc and gs from P-M, L15, and L15 without hourly values greater than 3 cm s−1 (“L15 gs low”) (cm s−1).
Error bars indicate two standard deviations (interannual spread). These quantities are calculated using observations during 1992-2000 (Section 3.2.3). (e) JJAS15 daytime mean P-M and L15 gs (cm s−1) driven by Harvard Forest observations for 1992-2014. Shades of red indicate years when Harvard Forest ozone EC observations are available (1992-2000);
other years are in shades of blue. Black indicates the multiyear means for 1992-2000 and 2001-2014. (f) JJAS15 daytime mean gs simulated by GFDL LM3 at Harvard Forest for 1990-2000 and L15 and P-M gs calculated with LM3 archived fields (Text B.3.2). For all quantities, the seasonal mean by hour has at least 25 days of data.
ozone dry deposition, whereas L15 gs yields an unphysical, negative gns. We thus use P-M to examine IAV in the relative stomatal contribution to canopy conductance (gs g−1c or
“stomatal fraction”) and gns (Figures 3.2c, B.3c,e).
From year to year, the JJAS15 daytime mean P-M-derived stomatal fraction spans 41-82% (multiyear mean is 60%) at Harvard Forest (Figure 3.2c). The summertime stomatal fraction over Kane Experimental Forest is 55% for 1997 (Zhang et al., 2006). Estimates for mixed temperate forests are 34% for 2002 (Hogg et al., 2007), 28% for 2000-2010 (Neirynck et al., 2012), 50% for 1998 (Zhang et al., 2006), and 47% for 2007 (Nunn et al., 2010).
JJAS15 mean gs accounts for a substantial fraction of gc each year at Harvard Forest, but the variance in P-M gs across years is 5% of the variance in daytime mean gc, in contrast to 106% for P-M-derived gns. The ranking across years of vdand gcis similar to that of gns (and inverse that of the stomatal fraction; Figure 3.2c), implying that gns drives IAV in vd. Meteorological and carbon and energy eddy covariance (EC) measurements have con-tinued after ozone EC was disconcon-tinued at Harvard Forest, allowing us to compute P-M and L15 for more recent years (Figures 3.2e, B.4). The relative interannual spreads in P-M and L15 gs estimates are ±26.9% and ±11.6% for 2001-2014. While the relative spread in L15 slightly decreases, the P-M relative spread for 2001-2014 is 2.5 times that for 1992-2000. This finding raises the possibility of decadal variability in the contribution of gs to IAV in gc. Our conclusion that gns drives IAV in JJAS15 mean vd thus may be limited to 1992-2000. Ozone dry deposition and coincident meteorological observations are needed for several decades to elucidate the impact of IAV in gs on gc.
We return to 1990-2000 to examine gs as simulated by GFDL LM3 at Harvard Forest.
The relative interannual spread of JJAS15 daytime mean LM3-simulated gs is ±18.3%, slightly higher but similar to that for the 1990-2000 observation-driven estimates. The ranking of years for LM3-simulated gs does not follow that for vd from ozone EC (Figure 3.2c,f), consistent with our conclusion from the observation-driven gsestimates that IAV in gsdoes not control IAV in vdat Harvard Forest during the 1990s. Similar to the observation-driven L15 gs, we find that the LM3-simulated gsis greater than gc, indicating unphysical, negative gns (Figure 3.2c,f). Even if we scale the LM3-simulated gs to the magnitude of the observation-driven P-M 1992-2000 JJAS15 daytime mean gs(Figure 3.2b), the variance
only explains 14% of the variance in gc across years, implying a dominant role for IAV in gns on gc.
We use LM3 to test how well the P-M and L15 approaches capture the IAV in LM3-simulated gs (Figures 3.2f, B.5). P-M and L15 gs are calculated from archived fields (Text B.3.2). Relative interannual spreads are similar, but slightly higher for LM3-driven L15 and P-M estimates (±22.5% and ±21.9%) than for LM3-simulated gs. The LM3-driven P-M and L15 approaches emulate the ranking across years in LM3-simulated gs (Figure 3.2f), indicating that both methods are capable of capturing IAV in LM3-simulated gs. This consistency across independent approaches lends confidence to constraining gs IAV using observation-driven P-M and L15.
The diurnal course of ecosystem-scale summertime mean gs at Harvard Forest remains uncertain. The 1992-2000 multiyear mean diurnal cycle calculated via P-M has a broad daytime maximum, whereas L15 simulates gs with an early-morning peak declining into the afternoon (Figures 3.2d, B.3b,d). This early-morning peak in observation-driven L15 gs also occurs in LM3-driven L15 (Figure B.5), but is produced neither by LM3 (Figure B.5) nor by P-M estimates (Figures B.5, 3.2d, B.3b). Leaf-level gs measurements during summers 1991-1992 (Bassow and Bazzaz, 1999) offer little constraint on the summertime mean diurnal cycle (Figure B.5). Differences (e.g., in IAV, magnitude, diurnal-cycle shape) between 1992-2000 observation-driven L15 and P-M lessen when we omit hourly gs greater than 3 cm s−1 from L15 (Figure 3.2d). For 84% of times that gs is greater than 3 cm s−1 during JJAS15 1992-2000 for 5am-6pm, VPD is less than 0.02 kPa, suggesting that some of the differences between P-M and L15 are from an overly-high sensitivity of gs to VPD in L15. Note that when L15 gs is greater than 3 cm s−1 during JJAS15 5am-6pm, mean GPP is 1.3 mol m−2 s−1 lower when VPD is less than 0.02 kPa. If stomatal and nonstomatal processes operate most strongly at different times of the day, constraining the stomatal contribution to the summertime mean diurnal cycle of vd should provide insight into driving nonstomatal mechanisms.